A hybrid automated detection of epileptic seizures in EEG records☆
Introduction
Epilepsy is a common long-term neurological condition in which nerve cell activity in the brain becomes disrupted. According to the WHO [1], about 0.6% of the general population suffers from epilepsy and nearly 80% of the affected people are found in developing regions. It manifests clearly through “epileptic seizure”, a period of sudden recurrent and transient disturbances of perception or behavior resulting from excessive synchronous activity of neurons in the brain. Only two third of epileptic patients can control seizures through medications. The remaining third develop drug-resistant epilepsy and are referred for a resection surgery where seizure onset zone (SOZ) is removed. Candidates for the surgery undergo many pre-surgery investigations where the most important is the long term Electroencephalographic (EEG) monitoring to capture seizures for off-line analysis. Expert neurologists then visually inspect the collected EEG data to detect epilepsy-related activity so they can determine the area where the seizures begin, called the seizure focus, and hence conclude whether the surgery is feasible [2]. The EEG has been a very useful clinical tool; it measures the electrical activity and represents potential differences from different sites of the human brain. EEG video monitoring generates a lengthy EEG recordings data for the clinical neurophysiologists to review and analyze seizures that occurred during the monitoring period. However, locating epileptic activity in a continuous EEG recording lasting several days or weeks is an exhausting, demanding and time-consuming task because such activity represents a small percentage of the entire recording. In an earlier study [3], that included 4 human experts to mark an EEG of 8 h, it was shown that here was little variation between readers. High seizure rate, short seizure duration and long seizure durations with ambiguous offsets can complicate the analysis and result in poor correlation. These difficulties have motivated the development of automated methods that scan, identify, and then present to a neurophysiologist epochs containing epileptic events. Such systems help to overcome the limitations of the traditional visual inspection performed by expert neurologists and avoid any misreading or missing information.
This paper introduces an automated seizure detection model that integrates Weighted Permutation Entropy (WPE) as input feature to a Support Vector Machine (SVM) learning model to enhance the sensitivity and precision of the detection process. WPE is a modified statistical parameter of the permutation entropy (PE) suggested by Bandt and Pompe [4]. It measures the complexity and irregularity of a time series by combining the ordinal pattern and the amplitude of its sample points. The feasibility of WPE as a feature for automated seizure detection has not been investigated so far.
The paper is organized as follows: Section 2 provides a cover of some related work presented in literature. Section 3 describes the medical background of epilepsy and brain seizures onto which the work was built, while Section 4 introduces the proposed seizure detection model and the dataset used. The model validation and experiments are demonstrated in Section 5. Results and discussion are illustrated in Sections 6 and 7, respectively. Finally, conclusion and suggestions for future work are illustrated in Section 8.
Section snippets
Literature survey
The automatic seizure detection system is a two stage problem; EEG recordings serve as input to a feature extraction phase, and then the features extracted are fed into a classifier. The features extraction methods exploited by researchers include; time-based features such as amplitude, duration, and sharpness [5]. Also frequency-based features such as fast Fourier transform [6], power spectral density [7]. The EEG signal being a non-stationary signal can be described by time-frequency features
Medical background
The role of EEG in the seizure detection process is very important compared to other approaches such as Magnetoencephalography (MEG) and functional Magnetic Resonance Imaging (fMRI) in terms of safety and cost. For an epilepsy diagnosed patient, the EEG recordings can belong to one of four categories; Pre-ictal refers to the state immediately before the actual seizure, though it has recently come to light that some of characteristics of this stage (such as visual auras) are actually the
Materials and methods
The proposed model includes a data acquisition phase followed by a pre-processing phase, a phase for segmenting each EEG channel into fixed length windows, a feature extraction phase where the Weighted Permutation Entropy (WPE) is estimated for each window, and finally a classification phase where EEG records are divided into one of two clinically significant EEG classes: (1) seizure and (2) no-seizure. The procedure of the proposed SVM classification for automated epileptic seizure detection
Model validation
Experiments have been carried out to validate the efficiency of the proposed seizure detection system and also its capability of detection and robustness to noise. Although the data set has five different classes of EEG segments (healthy with eyes open, healthy with eyes closed, interictal state obtained from epileptic zone, interictal state obtained from hippocampus zone of brain, and ictal state). The aim of this study is to detect segments with epileptic seizure activity so the multi-class
Classification of raw EEG signals
Fig. 7(a) shows the corresponding estimated WPE values (τ = 1, m = 3) for the entire 23.6 s EEG segments buffered in non-overlapping technique depicted in Fig. 2. The figure shows that the estimated WPE values are the lowest for EEG seizure segments (set S) compared to all the other datasets. A single Weighted Permutation Entropy value is calculated per window. For the overlapping segmentation techniques, same patterns are obtained that can be observed in Fig. 7(b). Furthermore, Fig. 8 shows
Discussion
An automated seizure detection model is introduced based on Weighted Permutation Entropy (WPE) and Support Vector Machine classifier (SVM). The experiments results show that the WPE values decrease during seizures which confirms the findings of previous studies that indicate that the brain activity during ictal state has a repetitive regular sequence and thus less chaos which yields to lower entropy values.
Comparing all obtained results, the best discrimination is obtained for seizure activity
Conclusion and future work
The proposed automated seizure detection model suggested scheme better tracks changes in the EEG signal through estimating the WPE value for each segment. Results show that the proposed model performed better than some existing schemes in literature according to the classification problem. The WPE algorithm was applied to raw and decomposed EEG signals. Another scenario was applied to simulate the real-time monitoring process, where the recording of EEG is accompanied by background noise
Noha Seddik Tawfik is teaching assistant at the Arab Academy for Science, Technology, and Maritime Transport, where she also obtained her M.Sc. and B.Sc degrees. Her research interests include biomedical engineering, pattern recognition and machine learning.
References (37)
Wavelet/mixture of experts network structure for EEG classification
Expert Syst Appl
(2008)- et al.
Epileptic seizure classification in EEG signals using second-order difference plot of intrinsic mode functions
Comput Methods Programs Biomed
(2014) - et al.
Classification of epileptic seizures in EEG signals based on phase space representation of intrinsic mode functions
Expert Syst Appl
(2015) - et al.
Classification of ictal and seizure-free EEG signals using fractional linear prediction
Biomed Signal Process Control
(2014) - et al.
Classification of seizure and seizure-free EEG signals using local binary patterns
Biomed Signal Process Control
(2015) - et al.
Automatic epileptic seizure detection in EEGs based on optimized sample entropy and extreme learning machine
J Neurosci Methods
(2012) - et al.
Detection of epileptic electroencephalogram based on permutation entropy and support vector machines
Expert Syst Appl
(2012) - et al.
Entropies for detection of epilepsy in EEG
Comput Methods Programs Biomed
(2005) - et al.
Automated diagnosis of epileptic EEG using entropies
Biomed Signal Process Control
(2012) ICTEDCT-CBIR: integrating curvelet transform with enhanced dominant colors extraction and texture analysis for efficient content-based image retrieval
Int J Comput Electr Eng
(2012)
Enhanced detection of artifacts in EEG data using higher-order statistics and independent component analysis
Neuroimage
Pharmacoresistant epilepsy: from pathogenesis to current and emerging therapies
Clevel Clin J Med
Seizure detection: correlation of human experts
Clin Neurophysiol
Permutation entropy: a natural complexity measure for time series
Phys Rev Lett
Detection of interictal spikes and artifactual data through orthogonal transformations
Clin Neurophysiol
Classification of epilepti-form EEG using a hybrid system based on decision tree classifier and fast fourier transform
Appl Math Comput
Automatic identification of epilepsy by HOS and power spectrum parameters using EEG signals: a comparative study
Cited by (121)
Landscape of epilepsy research: Analysis and future trajectory
2024, Interdisciplinary Neurosurgery: Advanced Techniques and Case ManagementSimplicial network analysis on EEG signals
2023, Physica A: Statistical Mechanics and its ApplicationsHierarchical Harris hawks optimization for epileptic seizure classification
2022, Computers in Biology and MedicineAutomated technique for EEG signal processing to detect seizure with optimized Variable Gaussian Filter and Fuzzy RBFELM classifier
2022, Biomedical Signal Processing and ControlEpileptic-seizure classification using phase-space representation of FBSE-EWT based EEG sub-band signals and ensemble learners
2022, Biomedical Signal Processing and ControlCitation Excerpt :Compared to other methods, the proposed method shows almost 3 % improvement in classification accuracy of classification problem C1, similarly 3.55 % in classification problem C2, and 2.1 % in C3. In recent works [53,54,63,64], authors have used DWT to decompose EEG signals. The major drawback of DWT is that it uses a pre-fixed function for analysing signals which makes it non-adaptive.
Noha Seddik Tawfik is teaching assistant at the Arab Academy for Science, Technology, and Maritime Transport, where she also obtained her M.Sc. and B.Sc degrees. Her research interests include biomedical engineering, pattern recognition and machine learning.
Sherin M. Youssef is professor, head of computer Engineering Department. She obtained her Ph.D., M.Phil. from University of Nottingham, UK (2004) in the field of intelligent systems. She did M.Sc. from University of Alexandria (1996) in machine learning & pattern analysis. Her research interest lies in the field of signal processing, video/image processing, image coding, intelligent systems, signal compression, biomedical engineering, biometrics, QoS intelligent networks, machine learning, genetic algorithms, neuro-based systems, swarm intelligence, multimedia systems.
Mohamed Kholief is an associate professor. He is the vice dean for academic and student affairs, College of Computing and Information Technology, at the Arab Academy for Science, Technology, and Maritime Transport in Alexandria, Egypt. He obtained his Ph.D. in computer science from Old Dominion University (Norfolk, Virginia) in 2003, and his M.Sc. and B.Sc. in computer science from Alexandria University (Egypt) in 1997 and 1991, respectively. His research interests include digital libraries, semantic web, e-learning and M-learning, software engineering, and knowledge Management.
- ☆
Reviews processed and recommended for publication to the Editor-in-Chief by Associate Editor Dr. M. R. Daliri.